Experimental Economics and Consumer Research

David Grether, California Institute of Technology
Louis Wilde, California Institute of Technology
ABSTRACT - Laboratory techniques have recently become accepted in economics. Experimental methodology in economics has developed along somewhat different lines than experimental methodology in consumer research. This paper presents an outline of the basic methodology of economic experiments and gives examples of recent experiments on individual choice behavior and on market outcomes
[ to cite ]:
David Grether and Louis Wilde (1984) ,"Experimental Economics and Consumer Research", in NA - Advances in Consumer Research Volume 11, eds. Thomas C. Kinnear, Provo, UT : Association for Consumer Research, Pages: 724-728.

Advances in Consumer Research Volume 11, 1984      Pages 724-728

EXPERIMENTAL ECONOMICS AND CONSUMER RESEARCH

David Grether, California Institute of Technology

Louis Wilde, California Institute of Technology

[Support of the National Science Foundation is gratefully acknowledged.]

ABSTRACT -

Laboratory techniques have recently become accepted in economics. Experimental methodology in economics has developed along somewhat different lines than experimental methodology in consumer research. This paper presents an outline of the basic methodology of economic experiments and gives examples of recent experiments on individual choice behavior and on market outcomes

I. INTRODUCTION

While experimental techniques have long been a staple of consumer research methodology, they have only recently been accepted in the mainstream economics literature. Partly this has been due to economists' traditional focus on market outcomes as opposed to individual behavior; many economists feel that experimental techniques are of little value to economic research unless they shed light on the nature of real-world institutions. In order to overcome this skepticism, economic experimentalists have taken great care to guarantee that laboratory micro-economies are real live economic systems in which real economic agents make real economic decisions. As a result, experimental methodology in economics has developed along somewhat different lines than experimental methodology in consumer research. It is the purpose of this paper to highlight these differences and provide some examples of recent economic experiments that are particularly relevant to consumer research.

Section II will outline the basic methodology of economic experiments as recently expounded by Vernon Smith [1982]. Section III will offer some examples of experiments related to individual behavior and Section IV will do the Same for experiments related to market outcomes. Section V will summarize the discussion

II. LABORATORY EXPERIMENTS IN ECONOMICS

The fundamental objective behind a laboratory experiment in economics is "to create a small-scale microeconomic environment in the laboratory where adequate control can be maintained and accurate measurement of relevant variables guaranteed" [Wilde, 1981, p. 138]. As Vernon Smith has pointed out, "control" is the key word in this statement. Moreover, he argues that "control over preferences as the most significant element distinguishing laboratory experiments from other methods of economic inquiry" Smith, 1982, p. 931]. One might also argue that it is the key element distinguishing experimental economics from experimental consumer research. Thus an economic experimentalist faces two initial tasks, to ensure that a genuine microeconomic environment has been created and to ensure that adequate control of individual preferences can be exercised. To a large extent, the concern with control has dominated the intellectual development of experimental economics. This concern derives, in turn, from a focus on institutions and theory testing, as opposed to the somewhat more practical focus of many consumer researcher experiments. We are not claiming one approach or the other is better, but simply seek to explicate that taken by economists.

The purpose of this section, then, is to make these notions precise and to present a set of sufficient conditions, due to Wilde and Smith, which guarantee that both initial tasks have been fulfilled. Our discussion here will be an informal summary of a more detailed technical exposition given in Smith [19821 (see also Wilde 11981]). Examples will be given in Sections III and IV.

Simply put, a microeconomic environment is composed primarily of two elements, a collection of individuals and an institutional setting. Two properties characterize the individuals; they are assumed to possess consistent preferences and to make decisions so as to maximize their own well-being. These decisions act through the institutional setting in order to determine final outcomes, or allocations. In a microeconomic experiment, a reward structure then relates monetary values to these final outcomes. The sufficient conditions for a valid microeconomic experiment relate primarily to this reward structure

Precept 1: Nonsatiation. "Given a costless choice between two alternatives, identical (i.e., equivalent) except that the first yields more of a reward medium (e.g., U. S. currency) than the second, the first will always be chosen (i.e., preferred) over the second, by an autonomous individual" [Smith, 1982, p. 931].

Precept 2: Saliency. "The reward earned by an individual is tied to decisions made by that individual" Wilde, 1981, p. 1403.

Precept 3: Dominance. "The reward structure dominates any subjective costs (or values) associated with participation in the activities of an experiment" [Smith, 1982, p. 934].

Precept 4: Privacy. "Subjects [are] uninformed of the overall pattern of the reward structure across individuals" [Wilde, 1981, p. 142].

Nonsatiation and saliency guarantee that a genuine microeconomic environment has been created; rewards are tied to final allocations which depend on subjects' decisions (saliency), and those decisions matter (nonsatiation). Dominance and privacy then guarantee that control can be exercised over the preferences "induced" by the reward structure.

What does this all mean in practical terms? Primarily, it implies that economic experiments have a kind of "starkness" compared to consumer research experiments. Since preferences are induced by monetary payoffs, product labels don't matter. In fact, they are undesirable as they may make control more difficult (dominance may become problematic). Secondly, it reflects a real concern with incentives, much more so than in most consumer research experiments -- if one is to be able to control preferences then one must be virtually paranoid about the incentives facing one s subjects. Most economists are extremely skeptical of any experiment which does not offer subjects rewards which are unambiguously desirable to them (e.g., cash) and somehow tied to the decisions they make in the experiment. Section III will present an example of the sort of problem that can arise when subjects are given hypothetical payoffs only. Finally, as mentioned above, there is a clear emphasis on institutional structure as well as individual behavior, although the same precepts apply to those economic experiments which do focus on individual behavior.

III. INDIVIDUAL BEHAVIOR

This section will summarize two recent economic experiments which focus on individual behavior. The first concerns the well-known conjunctive choice strategy and the second concerns individual's ability to implement Bayes' Rule. We will also discuss incentive effects (i.e., the importance of monetary rewards) in the context of the latter experiments. We will, however, given the objectives of this paper, concentrate more on methodology than results.

a. Conjunctive Choice

The conjunctive choice rule is a well-known strategy for selection among multi-attribute alternatives. Wright (1975) specifies it as requiring the decisionmaker to set up minimum cutoff levels on each attribute of the choice item. In deciding whether to accept an element of the choice set, the decisionmaker inspects each attribute of the element and accepts it only if it is above the cutoff level on every attribute. Failure to meet the preset standard for any attribute leads to rejection of the element under consideration and means the entire inspection process must start over with a new element from the choice set.

A set of experiments designed to test individuals' ability to implement the conjunctive choice rule was recently conducted by Grether and Wilde (1983). The three key aspects of this experiment were a payoff function (linking acceptable elements of the choice set to monetary rewards) expectations (regarding possible levels of attributes), and inspection costs (which must be paid when any attribute of any element of the choice set is observed ) . It was necessary to be able to control all three of these aspects. hence instead of using realworld products in a hypothetical setting, these authors used abstract products and cash payments.

In these experiments a "product" was described by a vector of real numbers, say (x1,x2,x3), for a good with three attributes. Payoffs were simply the sum of the realized value of these attributes, e.g., x1 + x2 + X3, less any inspection costs incurred. Expectations were specified by using uniform distributions for each attribute and informing subjects of the range of possible values. Inspection costs, which could differ across attributes, were also stated in dollar terms. The subject's task was to provide a set of cutoff levels for each of several problems.

During the course of the experiment, subjects were given a total of seventeen separate conjunctive choice problems, three one-attribute, six two-attribute, and eight three-attribute. All random numbers used during the experiment were generated "on the spot" in order to minimize any suspicious the subjects might have had of being unfairly manipulated. Since this was a time-consuming process, only one of the seventeen problems was used to generate actual payoffs. Subjects were informed of this at the beginning of the experiment, and that the problem on which actual earnings were to be based would be selected at random (using a bingo cage) by a monitor (chosen by the subjects). Again, the attribute values were drawn from uniform distributions, thus facilitating the use of a bingo cage to generate values for these random numbers too.

Terms such as "utility," "subjective distributions," "uniform," etc. were never used during any phase of the experiments. Subjects were shown objectively how the values of attributes were to be determined (using the bingo cage), told that their task was to set cutoff levels, and that their payment would be the value of the summed attributes for the first acceptable item less any inspection costs incurred in finding such an item.

Initially, subjects were given written instructions covering the one-attribute problem only. After reading the instructions to the subjects and answering any questions, an example showing how to calculate payoffs was done on a blackboard. Once subjects appeared to understand the task they each chose cutoffs for a one-attribute problem. When finished, random numbers were generated and each subject computed what his or her earnings would be from this problem if it happened to be the one that determined payment. Subjects then chose cutoffs for two additional one-attribute problems.

Procedures for the two-attribute problems were identical to the one-attribute problem, including generating the level of attributes randomly for the first problem and calculating earnings. Actually generating the random draws gave the subjects some experience and feedback and also insured that they understood the task confronting them, at least at an operational level. Subjects were told that numbers would be drawn until they had at least equaled their cutoffs on the first attribute. Then a single draw would be made for the second attribute. If the number exceeded their cutoff on the second attribute, they were done and should calculate their earnings. Otherwise, they would have to start over beginning with the first attribute. This procedure was operationally equivalent to drawing pairs of numbers and inspecting them sequentially.

After the two-attribute problems were completed, the three-attribute problems were introduced. Random numbers were not generated for any of the three-attribute problems unless one was selected to determine earnings, as the two-attribute problems familiarized subjects with the general form of the multi-attribute case.

These various design features can be seen to relate directly to the four precepts discussed in Section II of this paper. Choice items are abstract vectors, meaningless in the absence of the reward structure which, in this case, associates a dollar payoff equal to the sum of the components minus total accumulated inspection costs for any given acceptable vector (one which exceeds the cutoff level on each component). Great efforts are made to ensure the dominance precept is satisfied as well. Privacy, in this case, is relatively trivial since there is no interaction among subjects.

The reader can consult Grether and Wilde (1983) for a detailed description of these experiments and the results obtained. Generally speaking, it was found that most individual's acted roughly in accord with predictions based on the assumptions of net wealth maximization and risk neutrality. In particular, when an inspection cost increased, other things equal, the cutoff level on that attribute fell while the cutoff level on other attributes remained constant.

b. Bayes Experiments

Economic models which explicitly incorporate uncertainty require assumptions as to how economic agents adjust to new information concerning their environment. Typically these models include assumptions that agents have coherent beliefs that are consistent with the probability calculus and that agents' subjective probabilities agree with objective probabilities. Recently several psychologists have presented evidence of systematic biases in actual individual probability judgments and have identified a number of rules of thumb or heuristics that people use in making judgments involving uncertainty (see Slovic and Tversky). The experiments described in the section were designed to test one of these heuristics viz representativeness.

This experiment was conducted using as subjects students from University of California at Los Angeles. Students were recruited from economics classes. They were told that an "economics experiment" was to be held, given the time and place of the experiment, told that the experiment would not last longer than one hour, and that the minimum payment would be five dollars. When subjects arrived, they were randomly sent to one of two different rooms. Procedures in the two rooms were identical except for method of payment. A total of ninety-nine people participated in the experiment.

Instructions were passed out and after obtaining the participants' names, social security numbers, and addresses, the first three paragraphs of the instructions were read out loud to the subjects.

At this time the subjects elected one of themselves to be a monitor. The monitor was allowed to inspect all equipment, and, more importantly, to observe all procedures during this experiment. It was stated in the instructions that the monitor "should check the truthfulness of what the experimenter says, but other than that may not communicate any information to you in any way. If the monitor communicates any other information, he or she will be asked to leave without payment."

Though subjects could watch the experiment being performed, it was hoped that the use of the monitor would further increase credibility. After the monitor was elected, the remainder of the instructions were read and subjects questions answered. Once it appeared that subjects understood what their tasks were, the procedures were gone through (i.e., a dry run so to speak) to make sure that the subjects fully understood the mechanics of the experiment.

The equipment consisted of four bingo cages and an opaque screen. One of the cages designated Cage X contained balls numbered 1 through 6, and another cage, Cage Y, contained balls numbered 1 through 50. The other two cages (Cage A and Cage B) each contained six balls marked with the letters "N" or "G" (four" s and two G's in Cage A and three" s and three G s in Cage B).

The experiment proceeded as follows: Cages A and B were ?ut behind the screen and the rule used to determine which of them was to be chosen was announced; Cage X served as the "prior." The rules were of the form "if one of the numbers one through k is drawn from Cage 2, we shall choose Cage A; otherwise we shall choose Cage B." k was varied between two and four, thereby generating prior odds ratios for Cage A of one-half, one, and two. Next, a ball was drawn from Cage X (the subjects could not see the number on the ball) and, depending upon that number, either Cage A or Cage B was chosen and a sample of size six was drawn (with replacement) from it. After each draw the result ("N" or "G") was announced, written on a blackboard, and the subjects also recorded it on forms provided. After the six draws were completed, subjects were asked to indicate on their forms "the one (cage) you think the balls came from."

Next subjects were given the opportunity to choose to play one of two "bets" called the "Cage Bet" and the "Number Bet." A subject won the Cage Bet if the cage the subject picket (A or B) was indeed the cage the balls were drawn from. For the number bet the experimenter draws two balls (with replacement) from Cage Y, and if the number on the second ball is less than or equal to the number on the first ball, the subject wins. Which bet the subject played was determined as follows: subjects circled a number from 0 to 50 on the form provided. If the number circled was less than the first number drawn from Cage Y, the subject played the number bet; otherwise the subject played the Cage Bet.

The procedures are a variant of the method of Gordon Becker, DeGroot, and Marschak (1964) for determining demand prices. Thus it is a dominant strategy for the subjects to reveal their true personal probabilities. In other words, there is no incentive for the subjects to lie or otherwise distort their reported probabilities. For example, if a subject's subjective probability for Cage A is between .6 and .62, the subject should indicate that Cage A is more likely and circle the number 30. If a subject's subjective probability is exactly .6, the subject should be indifferent between circling 30 and circling 29.

Suppose that a subject s subjective probability is .59, but that the subject circles a number other than 29. If the number circled is less than 29, then the subject runs the risk of playing the number bet with a probability of winning that is less than .59. Similarly, if a subject circles a number greater than 29, say 38, then if numbers 30 through 38 are drawn, the subject will play the cage bet and miss the chance to play the number bet at odds more favorable than .59/.41.

In one room all subjects were paid $7 for participating. In the other room subjects were told that at the end of the experiments one of their decisions would be randomly selected (using a bingo cage) and that those whose decisions were correct, i.e., those that won the relevant bet (Cage Bet or Number Bet) would receive $15, otherwise $5.

While preparing the data on subjects responses, the circled numbers were converted into estimates of the subjects' personal probabilities. Note that given the procedures subjects should never respond by circling numbers 0 through 23 and only those subjects whose personal odds on the cage bet are exactly one to one could reasonably respond by circling 24. This assumes, of course, that subjects would rather win their bets than lose them.

Some subjects did in fact give such responses and, the frequency of occurrences was greater for those without financial incentives (126 out of 980) than for those with financial incentives (42 of 960 responses). The difference is statistically significant at any conventional level of significance (X (l) = 44.1).

Notice that this experiment was designed to control the incentives and belief of the subjects. The use of a monitor, physical randomizing devices, and cash payments all combine to ensure the four precepts. Detailed results are presented; Grether (1980, 1981).

IV. MARKET EXPERIMENTS

In this section we will describe two sets of experiments related to market behavior. The first concerns several models of the effects of shopping behavior on market equilibria and the second concerns the so-called "lemons effect."

a. Shopping Models

The recent economics literature has shown a great deal of interest in models that relate various shopping strategies to market equilibria. Three of these interest us here. All make similar assumptions regarding firms -each firm faces an identical technology described by a fixed cost, a constant marginal cost and a capacity constraint. Each produces an identical homogeneous product and maximizes expected profits. The equilibrium is given by a number of firms (determined by a zero-profit, free-entry condition) and a distribution of prices (one for each firm) such that each firm maximizes its profits taking the other firms prices as given.

Consumers all demand one unit of the product or none. Each is willing to pay up to some exogenous "limit price" to obtain the good. They differ only according to their shopping strategies. The three models are as follows:

(i) Schwartz-Wilde (1979): Buyers are divided into two groups, shoppers and nonshoppers. Shoppers sample some fixed number of firms and buy from the lowest-priced. Nonshoppers sample one firm at random.

(ii) Salop-Stiglitz (1977): Consumers have the opportunity to purchase information, at some cost, which gives them a list of all prices. Thus for a fee buyers may purchase from the firm with the lowest price. If they choose not to buy this information they shop one firm at random, for free.

(iii) The Monopoly Model: This model is similar to Salop-Stiglitz except that the information source provides only a sample of prices, chosen randomly, instead of all prices.

The Schwartz-Wilde model predicts either a competitive equilibrium or an equilibrium with "smooth" price dispersion above the competitive price, depending on the mix of shoppers and nonshopper ;. The Salop-Stiglitz model, under the experimental parameters, predicts a two-price equilibrium, with some firms charging the competitive price and some the limit price. The monopoly model, clearly, predicts all firms charging the limit price. The following experiment was designed to test these models (Schwartz, Grether and Wilde, 1983).

The experiments were conducted on the campus of California Institute of Technology using Caltech undergraduates as subjects. The subjects were recruited for economics experiments, they were told that they would be paid in cash at the end of each experiment but were not told of the nature of the experiments.

The experimental sessions were all organized in basically the same way. Subjects were divided into two groups -buyers and sellers -- and given sets of instructions. These were read aloud to all subjects at the beginning of the session. Participants were given their own parameters for the market but were not told the values of the parameters for the other participants. After the instructions were read numerical examples were presented.

Sellers were told that they could make money by selling units of a commodity to the buyers. They could buy the units from the experimenters at preset prices and could keep all profits from these trades (plus a $.10 commission). The cost schedules for the sellers included a fixed cost per period (called a participation fee) and a constant marginal cost for each unit ordered up to a fixed limit (capacity). Sellers did not have to participate and could, thus, avoid paying the fixed costs. However, sellers that did not participate could not buy or sell units or be active in the experiment in any way. This is the way that the free entry and exit conditions of the models were implemented Sellers were prohibited from selling any units below the competitive price. The purpose of this was to limit the possible losses sellers could incur. In an experiment of this type subjects typically (and correctly) do not believe that they will be made to pay if at the end of the experiment they have an overall loss. Therefore in designing the experiment care has to be taken to control for this eventuality. For these experiments this was done by giving each seller a lump sum to which all profits and losses were added. The sum varied from $10 to S]5 depending on the cost parameters. The price floor was a way of ensuring that the experimenters did not face the credibility problem that losses would have caused.

Sellers were seated facing a blackboard. At the beginning of each market round sellers would decide whether or not to participate and, if participating, what price to set for that period. Prices were collected from the participating sellers and posted on the blackboard (though the identities of the sellers were not given). After the buyers had placed their orders these were also posted on the blackboard so that during the experiment the sellers could see the complete history of the prices charged and the volumes sold at each price. Each participating seller was told the number of units he or she had sold. At this time they had to decide on how many units to order, to record these transactions, and to update their inventory records. When all sellers were finished with these tasks another market period was begun. At the end of the session all stocks in inventory were lost; that is the experimenters would not redeem them for cash. If a seller had unfilled orders the seller had to buy the units for them from the experimenters at the same price as it would have cost to "produce" them during the experiment including paying the fixed cost once for each multiple of their capacity or fraction thereof needed.

The buyers were seated facing another blackboard which the sellers could not see. When the experimenters had picket up the seller prices at the start of a period they were posted on the buyers' blackboard with or without seller numbers depending upon the model being tested. At the beginning of an experiment buyers were given transaction record sheets to record their purchases. These sheets had the seller numbers of the sellers that they could buy their units from written on them in advance. For some of the experiments these preassigned numbers were covered up so that the buyers would not know which sellers they were going to deal with. The reason for this procedure was that for some of the models buyers had to decide between going to a single seller chosen at random or paying a fee and then engaging in some form of shopping behavior. Buyers often kept complete price histories on all sellers during an experiment and would use this information in deciding whether to go to the preassigned seller or to pay the fee and "shop." After the buyers had made their purchases they were recorded by the experimenters, the sellers were individually informed as to their sales and the volume data posted on the sellers' blackboard.

Buyers were told that they could buy units of the good from the sellers and resell them to the experimenters. They could keep all profits from these transactions plus, again, a ten cent commission. The experimenters would buy the units from the buyers at a fixed price per unit and buyers were not allowed to buy units priced above this price (termed the redemption value). Buyers were classified in two groups -- shoppers and nonshoppers. For all the experiments shoppers would go to a single seller (preassigned) and purchase a single unit from that seller provided the price was less than their redemption value. Shoppers sampled from a set of sellers, buying from the lowest priced firm sampled provided, again, that this price was less than the redemption value. The nature of the sample of firms and the decision as to whether or not to shop were the main control variables for this set of experiments.

For the experiments testing the Schwartz-Wilde model the buyers were essentially passive. That is, the shoppers were given previously determined samples of sellers and simply purchased one unit from the lowest priced seller if that price was one at which they could buy. The procedures for the Salop and Stiglitz models were quite different. After the sellers had submitted their prices the complete price distribution was posted on a blackboard that only the buyers could see. Then each buyer had to decide whether to be a shopper or a non hopper. In order to be a shopper a buyer had to pay a fee (either $.10 or $.30 per unit) which entitled the buyer to buy one unit from the lowest priced seller. In each round there were 9 units with the smaller fee and 16 units that required the larger one. Experiments using the monopoly model had a similar buyer technology. In these experiments the buyers could either buy from a single seller or for a fee they could obtain a sample of size two and purchase from the lower priced one.

Surprisingly, all three models yielded experimental outcomes consistent with the theoretical predictions. However, our interest here is once again on the methodological features. In particular, we note that demand and supply curves were "induced" by using "redemption values" and "unit costs," respectively, which reflected actual cash values. Furthermore, a "commission" was included as part of the reward for each transaction. This is another example of the economic experimentalist's concern with incentives. Experience has shown that subjects will often not make "marginal trades" (ones for which they earn no profit) unless such a "commission" is included.

b. Lemons Model

The development of differentiated products in response to market competition and market organization were studied in Lynch, Miller, Plott, and Porter (1983). In this study an underlying commodity was designated as being of two possible grades, Regular or Super Buyers who purchased a Super could redeem it with the experimenter for more money than could buyers who purchased Regulars. Other things equal, buyers preferred Supers. Each seller was constrained by the experimental design to supply no more than a fixed number of units to the market, and units supplied were done so only at a cost to the seller. Subject to these constraints, each seller was free to choose the combinations of Regulars and Supers in his/her supply and the total number of units supplied. The cost to the seller of supplying Supers was greater than the cost of Regulars so, other things being equal, sellers would prefer to supply Regulars.

Notice that minor changes in the basic design of the previous section -- alteration of the cost schedules of the sellers and the redemption schedules of the buyers have been used to produce heterogeneous goods viz Super and Regulars. In the initial experimental setting buyers had no way to determine which units offered for sale were Super and which were not. In this initial set up it was not possible to identify a unit with a given seller. Under these conditions sellers do not have sufficient incentives to produce the high quality goods (Super) even though buyers are willing to pay enough for them to make their production profitable. By tagging goods with seller identities (brand names) rival "brands" can be introduced and sellers can invest in building up a reputation as high (low) quality producers if they so choose. Varying the sales contract terms e.g. a minimum redemption value or the seller refunds part of the purchase price introduces other aspects commonly found in the market Place .

Warranties, sellers reputations, the cost and effectiveness of reputation development, etc. were studied. The purpose of this study, which was financed by the FTC, was to provide a series of background experiments against which the FTC could study proposed policies regarding product quality and reliability and the claims frequently heard that under some conditions competition can act perversely in generating good quality products.

V. CONCLUSION

The emphasis in experimental economics has been on theory testing and on studying the effects of market institutions or econometric outcomes. Economists have adopted procedures with major emphasis on the incentives structures facing the subjects and with stark environments. Note, for example, the way in which "brands," different qualities of goods, "advertising," money back guarantees, etc. were implemented in the study of the lemons markets. The aim was to set up a real market (in the laboratory) designed to exhibit certain features of natural (i.e., nonlaboratory) markets, but not necessarily to simulate the market for any particular product such as used automobiles, TV dinners, homes, or personal services.

The examples of the previous two sections illustrate the nature of experimental methods in economics both for individual choice experiments and for market experiments. They provide examples of the role of the four precepts nonsatiation, salience, dominance, and privacy discussed in Section II.

REFERENCES

Becker, Gordon M.; DeGroot, M. B. and Marschak, J. (1964), "Measuring Utility by a Single-Response Sequential Method," Behavioral Science, 9, 226-32.

Grether, David (1980), "Bays Rule as a Descriptive Model: The Representativeness Heuristic," Quarterly Journal of Economics, 95, 537-57.

Grether, David (1981), "Financial Incentive Effects and Individual Decision-making," California Institute of Technology Social Science Working Paper No. 401.

Grether, David and Wilde, Louis (1983), "An Analysis of Conjunctive Choice: Theory and Experiments," forthcoming in J. Consumer Research (M.arch 1984).

Kahneman, Daniel; Slovic, Paul and Tversky, Amos (1982), Judgment Under Uncertainty: Heuristics and Biases, New York: Cambridge University Press.

Lynch, M.; Miller, R.; Plott, C. and Porter, R. (1983), "Asymmetric Information, Adverse Selection, and Lemons in Experimental Markets." Paper presented at the Western Economic Association. California Institute of Technology, mimeographed.

Plott, C. R. (1982), "Industrial Organization Theory and Experimental Economics," Journal of Economic Literature, 20. 1485-1527.

Salop, Steven and Stiglitz, Joseph (1977), "Bargains and Payoffs: A Model of Monopolistically Competitive Price Dispersion," Rev Econ Studies, 44. 492-510.

Schwartz, Alan; Grether, David; and Wilde, Louis (1983), "Equilibrium Search Models: An Experimental Test of Three Theories," California Institute of Technology, mimeographed.

Schwartz, Alan and Wilde, Louis (1979), "Equilibrium Comparison Shopping." Rev Econ Studies. 46. 543-553.

Smith, Vernon (1982), '*Microeconomic Systems as an Experimental Science," Ameri Econ Rev, 72, 923-55.

Wilde, Louis (1991), "On the Use of Laboratory Experiments in Economics," in Joseph Pitt, ed., Philosophy of Economics (Pordrecht, Holland: D. Reidel) pp. 137-48.

Wright, Peter (1975), "Consumer Choice Strategies: Simplifying vs. Optimizing," J. Marketing Research, 12, 60-67.

----------------------------------------